The Prediction Model Of Protein Subcellular Localization Based On Graphical Representation

The graphical representation of biological sequences is a valid method in bioinformatics for the visibility and mathematical descriptor. In the paper, two graphical representation of protein are introduced to compare the similarities of protein and predict subcellular location.A new graphical representation is suggested based on triplet codon, hydrophobic index of amino acid, and different parameters iterated function system. The mathematical descriptor is suggested to characterize the similarities/dissimilarities of two protein sequences. The usefulness of this approach can be illustrated by performing the comparison of sequences of ND5 proteins of nine species, as well as sequences of twelve β-globin protein. Based on the distance matrix,we construct their phylogenetic tree, in which consistent with the evolution of species in biology.By the correlation analysis, ClustalW results were compared with our results and some other graphical representation results to demonstrate the effectiveness of our approach in the study of sequence similarity analysis.Subcellular localization prediction is a hot issue in bioinformatics. In the paper, a novel subcellular localization prediction model is proposed based on graphical representation and BP neural network method. Firstly, by a novel graphical representation and mathematical descriptor,the distance matrix of protein sequences is computed. A new subcellular location prediction model is introduced based on the standardized distance matrix and the BP neural network.Applying the model to two data sets ZD98 and CL317, their overall prediction accuracy are94.9% and 87.4% respectively. The results of the individual sensitivity and overall prediction accuracy in our prediction method and some other method are compared to illustrate the efficiency of our approach in the subcellular location prediction.